Skip to content

patches on 1.70 #356

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
AtlantaPepsi wants to merge 5 commits into from
Closed

patches on 1.70 #356

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
f814206
removing dependency on ppa -> gpg-agent, which is absent on this img
AtlantaPepsi Sep 3, 2026
db4f461
Update EnvVars.hpp
AtlantaPepsi Sep 4, 2026
7eb0a8d
pingpong latency test
AtlantaPepsi Sep 8, 2026
c38de78
adjusting output format
AtlantaPepsi Sep 8, 2026
b568783
relaxed atomics for pingpong kernel
AtlantaPepsi Sep 9, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 1 addition & 4 deletions .github/workflows/codeql.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,11 +30,8 @@ jobs:
- name: Install OS requirements
timeout-minutes: 10
run: |
apt update
apt install -y software-properties-common
apt-add-repository -y ppa:git-core/ppa
apt-get update
apt install -y git
DEBIAN_FRONTEND=noninteractive apt-get install -y git

- name: Checkout repository
uses: actions/checkout@v4
Expand Down
24 changes: 16 additions & 8 deletions src/client/Client.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -107,16 +107,24 @@ int main(int argc, char **argv)
if (transfers.empty()) {
Print("<none>\n");
} else {
bool isMultiNode = GetNumRanks() > 1;
for (size_t i = 0; i < transfers.size(); i++) {
Transfer const& t = transfers[i];
Print("Transfer %5lu: (%s->", i, MemDevicesToStr(t.srcs).c_str());
if (isMultiNode) Print("R%d", t.exeDevice.exeRank);
Print("%c%d", ExeTypeStr[t.exeDevice.exeType], t.exeDevice.exeIndex);
if (t.exeDevice.exeSlot) Print("%c", 'A' + t.exeDevice.exeSlot);
if (t.exeSubIndex != -1) Print(".%d", t.exeSubIndex);
if (t.exeSubSlot != 0) Print("%c", 'A' + t.exeSubSlot);
Print("->%s)\n", MemDevicesToStr(t.dsts).c_str());
if (t.numLaps > 0) {
Print("Transfer %5lu: PingPong +%d: (%s->%s->%s) <+> (%s->%s->%s)\n",
i, t.numLaps,
MemDeviceToStr(t.srcs[0]).c_str(),
ExeDeviceToStr(t.exeDevice, t.exeSubIndex, t.exeSubSlot).c_str(),
MemDeviceToStr(t.dsts[0]).c_str(),
MemDeviceToStr(t.srcs[1]).c_str(),
ExeDeviceToStr(t.exeDevicePong, t.exeSubIndexPong, t.exeSubSlotPong).c_str(),
MemDeviceToStr(t.dsts[1]).c_str());
} else {
Print("Transfer %5lu: (%s->%s->%s)\n",
i,
MemDevicesToStr(t.srcs).c_str(),
ExeDeviceToStr(t.exeDevice, t.exeSubIndex, t.exeSubSlot).c_str(),
MemDevicesToStr(t.dsts).c_str());
}
}
}
return 0;
Expand Down
110 changes: 61 additions & 49 deletions src/client/EnvVars.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,7 +44,7 @@ THE SOFTWARE.
#include <random>
#include <time.h>

#define CLIENT_VERSION "01"
#define CLIENT_VERSION "00"

#include "TransferBench.hpp"
using namespace TransferBench;
Expand Down Expand Up @@ -74,6 +74,8 @@ class EnvVars
int numIterations; // Number of timed iterations to perform. If negative, run for -numIterations seconds instead
int numSubIterations; // Number of subiterations to perform
int numWarmups; // Number of un-timed warmup iterations to perform
int pingpongFlagBuffer; // Size in bytes of the pingpong flag buffer (must be positive)
int pingpongStride; // Stride in bytes between flag slots for pingpong laps (wraps within the flag buffer)
int showBorders; // Show ASCII box-drawing characaters in tables
int showIterations; // Show per-iteration timing info
int useInteractive; // Pause for user-input before starting transfer loop
Expand Down Expand Up @@ -158,55 +160,57 @@ class EnvVars
else if (archName == "gfx950") defaultGfxUnroll = 4;
else if (archName == "gfx1250") defaultGfxUnroll = 32;

alwaysValidate = GetEnvVar("ALWAYS_VALIDATE", 0);
blockBytes = GetEnvVar("BLOCK_BYTES" , 256);
byteOffset = GetEnvVar("BYTE_OFFSET" , 0);
fillCompress = GetEnvVarArray("FILL_COMPRESS" , {});
gfxBlockOrder = GetEnvVar("GFX_BLOCK_ORDER" , 0);
gfxBlockSize = GetEnvVar("GFX_BLOCK_SIZE" , 256);
gfxKernel = GetEnvVar("GFX_KERNEL" , 0);
gfxSeType = GetEnvVar("GFX_SE_TYPE" , 0);
gfxSingleTeam = GetEnvVar("GFX_SINGLE_TEAM" , 0);
gfxTemporal = GetEnvVar("GFX_TEMPORAL" , 0);
gfxUnroll = GetEnvVar("GFX_UNROLL" , defaultGfxUnroll);
gfxWaveOrder = GetEnvVar("GFX_WAVE_ORDER" , 0);
gfxWordSize = GetEnvVar("GFX_WORD_SIZE" , 4);
hideEnv = GetEnvVar("HIDE_ENV" , 0);
minNumVarSubExec = GetEnvVar("MIN_VAR_SUBEXEC" , 1);
maxNumVarSubExec = GetEnvVar("MAX_VAR_SUBEXEC" , 0);
numIterations = GetEnvVar("NUM_ITERATIONS" , 10);
numSubIterations = GetEnvVar("NUM_SUBITERATIONS" , 1);
numWarmups = GetEnvVar("NUM_WARMUPS" , 3);
outputToCsv = GetEnvVar("OUTPUT_TO_CSV" , 0);
samplingFactor = GetEnvVar("SAMPLING_FACTOR" , 1);
showBorders = GetEnvVar("SHOW_BORDERS" , 1);
showIterations = GetEnvVar("SHOW_ITERATIONS" , 0);
showPercentiles = GetEnvVarArray("SHOW_PERCENTILES", {});
sweepMaxPow2 = GetEnvVar("SWEEP_MAX_POW2" , 29);
sweepMinPow2 = GetEnvVar("SWEEP_MIN_POW2" , 10);
sweepMinPow2 = std::clamp(sweepMinPow2, 0, 62);
sweepMaxPow2 = std::clamp(sweepMaxPow2, 0, 62);
alwaysValidate = GetEnvVar("ALWAYS_VALIDATE", 0);
blockBytes = GetEnvVar("BLOCK_BYTES" , 256);
byteOffset = GetEnvVar("BYTE_OFFSET" , 0);
fillCompress = GetEnvVarArray("FILL_COMPRESS" , {});
gfxBlockOrder = GetEnvVar("GFX_BLOCK_ORDER" , 0);
gfxBlockSize = GetEnvVar("GFX_BLOCK_SIZE" , 256);
gfxKernel = GetEnvVar("GFX_KERNEL" , 0);
gfxSeType = GetEnvVar("GFX_SE_TYPE" , 0);
gfxSingleTeam = GetEnvVar("GFX_SINGLE_TEAM" , 0);
gfxTemporal = GetEnvVar("GFX_TEMPORAL" , 0);
gfxUnroll = GetEnvVar("GFX_UNROLL" , defaultGfxUnroll);
gfxWaveOrder = GetEnvVar("GFX_WAVE_ORDER" , 0);
gfxWordSize = GetEnvVar("GFX_WORD_SIZE" , 4);
hideEnv = GetEnvVar("HIDE_ENV" , 0);
minNumVarSubExec = GetEnvVar("MIN_VAR_SUBEXEC" , 1);
maxNumVarSubExec = GetEnvVar("MAX_VAR_SUBEXEC" , 0);
numIterations = GetEnvVar("NUM_ITERATIONS" , 10);
numSubIterations = GetEnvVar("NUM_SUBITERATIONS" , 1);
numWarmups = GetEnvVar("NUM_WARMUPS" , 3);
pingpongFlagBuffer = GetEnvVar("PINGPONG_FLAG_BUFFER", 1);
pingpongStride = GetEnvVar("PINGPONG_STRIDE" , 1);
outputToCsv = GetEnvVar("OUTPUT_TO_CSV" , 0);
samplingFactor = GetEnvVar("SAMPLING_FACTOR" , 1);
showBorders = GetEnvVar("SHOW_BORDERS" , 1);
showIterations = GetEnvVar("SHOW_ITERATIONS" , 0);
showPercentiles = GetEnvVarArray("SHOW_PERCENTILES", {});
sweepMaxPow2 = GetEnvVar("SWEEP_MAX_POW2" , 29);
sweepMinPow2 = GetEnvVar("SWEEP_MIN_POW2" , 10);
sweepMinPow2 = std::clamp(sweepMinPow2, 0, 62);
sweepMaxPow2 = std::clamp(sweepMaxPow2, 0, 62);
if (sweepMinPow2 > sweepMaxPow2) std::swap(sweepMinPow2, sweepMaxPow2);
tdmBlockOrder = GetEnvVar("TDM_BLOCK_ORDER" , 0);
tdmBlockSize = GetEnvVar("TDM_BLOCK_SIZE" , 256);
tdmLdsBytes = GetEnvVar("TDM_LDS_BYTES" , 0);
useHipEvents = GetEnvVar("USE_HIP_EVENTS" , 1);
useHsaDma = GetEnvVar("USE_HSA_DMA" , 0);
useInteractive = GetEnvVar("USE_INTERACTIVE" , 0);
useSingleStream = GetEnvVar("USE_SINGLE_STREAM" , 1);
validateDirect = GetEnvVar("VALIDATE_DIRECT" , 0);
validateOnDevice = GetEnvVar("VALIDATE_ON_DEVICE" , 0);
validateSource = GetEnvVar("VALIDATE_SOURCE" , 0);

ibGidIndex = GetEnvVar("IB_GID_INDEX" ,-1);
ibPort = GetEnvVar("IB_PORT_NUMBER" , 1);
roceVersion = GetEnvVar("ROCE_VERSION" , 2);
ipAddressFamily = GetEnvVar("IP_ADDRESS_FAMILY" , 4);
nicChunkBytes = GetEnvVar("NIC_CHUNK_BYTES" , 1073741824);
nicCqPollBatch = GetEnvVar("NIC_CQ_POLL_BATCH" , 4);
nicRelaxedOrder = GetEnvVar("NIC_RELAX_ORDER" , 1);
nicServiceLevel = GetEnvVar("NIC_SERVICE_LEVEL" , 0);
nicTrafficClass = GetEnvVar("NIC_TRAFFIC_CLASS" , 0);
tdmBlockOrder = GetEnvVar("TDM_BLOCK_ORDER" , 0);
tdmBlockSize = GetEnvVar("TDM_BLOCK_SIZE" , 256);
tdmLdsBytes = GetEnvVar("TDM_LDS_BYTES" , 0);
useHipEvents = GetEnvVar("USE_HIP_EVENTS" , 1);
useHsaDma = GetEnvVar("USE_HSA_DMA" , 0);
useInteractive = GetEnvVar("USE_INTERACTIVE" , 0);
useSingleStream = GetEnvVar("USE_SINGLE_STREAM" , 1);
validateDirect = GetEnvVar("VALIDATE_DIRECT" , 0);
validateOnDevice = GetEnvVar("VALIDATE_ON_DEVICE" , 0);
validateSource = GetEnvVar("VALIDATE_SOURCE" , 0);

ibGidIndex = GetEnvVar("IB_GID_INDEX" ,-1);
ibPort = GetEnvVar("IB_PORT_NUMBER" , 1);
roceVersion = GetEnvVar("ROCE_VERSION" , 2);
ipAddressFamily = GetEnvVar("IP_ADDRESS_FAMILY" , 4);
nicChunkBytes = GetEnvVar("NIC_CHUNK_BYTES" , 1073741824);
nicCqPollBatch = GetEnvVar("NIC_CQ_POLL_BATCH" , 4);
nicRelaxedOrder = GetEnvVar("NIC_RELAX_ORDER" , 1);
nicServiceLevel = GetEnvVar("NIC_SERVICE_LEVEL" , 0);
nicTrafficClass = GetEnvVar("NIC_TRAFFIC_CLASS" , 0);

// Check that NIC service level and traffic class are in valid ranges
if (nicServiceLevel < 0 || nicServiceLevel > 15) {
Expand Down Expand Up @@ -404,6 +408,8 @@ class EnvVars
printf(" NUM_SUBITERATIONS - # of sub-iterations to run per iteration. Must be non-negative\n");
printf(" NUM_WARMUPS - # of untimed warmup iterations per test\n");
printf(" OUTPUT_TO_CSV - Outputs to CSV format if set\n");
printf(" PINGPONG_FLAG_BUFFER - Size in bytes of the pingpong flag buffer (default 1, must be positive). Each flag slot is 1 byte\n");
printf(" PINGPONG_STRIDE - Stride in bytes between flag slots for pingpong laps (default 1, must be positive unless the flag buffer is 1 byte, wraps within the flag buffer)\n");
printf(" SAMPLING_FACTOR - Add this many samples (when possible) between powers of 2 when auto-generating data sizes\n");
printf(" SHOW_BORDERS - Show ASCII box-drawing characters in tables\n");
printf(" SHOW_ITERATIONS - Show per-iteration timing info\n");
Expand Down Expand Up @@ -555,6 +561,10 @@ class EnvVars
"Running %s subiterations", (numSubIterations == 0 ? "infinite" : std::to_string(numSubIterations)).c_str());
Print("NUM_WARMUPS", numWarmups,
"Running %d warmup iteration(s) per Test", numWarmups);
Print("PINGPONG_FLAG_BUFFER", pingpongFlagBuffer,
"Pingpong flag buffer of %d bytes", pingpongFlagBuffer);
Print("PINGPONG_STRIDE", pingpongStride,
"Pingpong flag stride %d bytes per lap", pingpongStride);
Print("SHOW_BORDERS", showBorders, "%s ASCII box-drawing characaters in tables", showBorders ? "Showing" : "Hiding");
Print("SHOW_ITERATIONS", showIterations,
"%s per-iteration timing", showIterations ? "Showing" : "Hiding");
Expand Down Expand Up @@ -738,6 +748,8 @@ class EnvVars
cfg.general.numIterations = numIterations;
cfg.general.numSubIterations = numSubIterations;
cfg.general.numWarmups = numWarmups;
cfg.general.pingpongFlagBuffer = pingpongFlagBuffer;
cfg.general.pingpongStride = pingpongStride;
cfg.general.recordPerIteration = ((showIterations != 0) || !showPercentiles.empty()) ? 1 : 0;
cfg.general.useHipEvents = useHipEvents;
cfg.general.useInteractive = useInteractive;
Expand Down
Loading
Loading